MCP servers
Plug external tools into the team, browsers, databases, APIs, and more.
Agent providers
Pick the premium brain that plans changes: Claude Code, Codex, and others.
What is an MCP server?
The Model Context Protocol (MCP) is an open standard that lets AI agents talk to external tools through a common interface. An MCP server exposes a set of capabilities, for example, driving a browser, querying a database, or calling a third-party API, that your agents can use during their work. When you connect an MCP server, every agent on the team gains access to its tools. The CEO orchestrator can delegate work that requires those tools, and specialists can call them while they plan.A popular example is the Playwright MCP server, which gives agents real browser automation, navigating pages, clicking elements, and reading the DOM to verify behavior.
Connect an MCP server
Configure the command
Provide the launch command and arguments for the server. For Playwright, point Orkestral at the
@playwright/mcp package so it starts on demand.Agent providers
Agent providers are the premium planning brains. They read your repos, reason about the change, and produce the plan and instructions. The bundled local Forge (Qwen2.5-Coder) then executes the code edits on your machine through the deterministic Fast Apply engine.Claude Code
Anthropic’s agent, strong at multi-step planning and large-codebase reasoning.
Codex
An alternative provider for planning and delegation across the team.
More providers can be added over time. Whichever you choose, the division of labor stays the same: the provider plans, the Forge executes locally.
Choose a provider
How integrations expand the team
Each integration adds a new capability that the whole hierarchy can draw on.More tools, broader work
More tools, broader work
Connecting an MCP server like Playwright lets QA agents test in a real browser, while Backend and DevOps agents can reach the APIs and services those servers expose.
Better planning
Better planning
Choosing a stronger agent provider improves how the CEO orchestrator decomposes your request into issues and delegates to specialists.
Local-first by default
Local-first by default
Integrations extend what the team can do, but execution stays on your machine through the Forge. Everything lives in
~/.orkestral, no server, no telemetry.Next steps
Meet the team
See the agents and the reporting hierarchy that use these integrations.
The Forge
Learn how local execution applies code at $0 API cost.